Across therapeutic areas, regulators and payers are tightening expectations that endpoints must reflect tangible patient benefit—how long patients live, how well they function, and how they feel—rather than relying solely on surrogate markers. Recent FDA guidance and methodological work highlight a dual track: (1) “hard” clinical endpoints (survival, functional status, major clinical events) as the gold standard, and (2) carefully validated surrogate or intermediate endpoints to enable accelerated or efficient development, especially in oncology and cellular/gene therapies (CGT) [1][2][3][4].

Key meaningful endpoints

1. Direct patient‑centric endpoints

These remain the anchor for most pivotal and confirmatory trials:

• Overall survival (OS)

• Defined as time from randomization to death from any cause.

• Explicitly identified as the safety and benefit–risk anchor in FDA’s 2025 oncology guidance: all randomized oncology trials should assess OS, at least as a key safety endpoint, even when not primary [2].

• Used as primary or key secondary endpoint in many oncology approvals; still viewed as the most interpretable, least assumption‑dependent endpoint for life‑threatening disease.

• Event‑free or progression‑based endpoints (by disease area)

• Progression‑free survival (PFS) or similar time‑to‑event measures in oncology (event‑driven, blinded assessment wherever possible).

• Major adverse clinical events in non‑oncology:

  • Nonfatal MI, stroke, cardiovascular death in cardiology.

  • Exacerbations/hospitalizations in respiratory disease.

  • Transplant‑free survival or decompensation events in organ failure trials.

• These remain meaningful when they are clearly linked to long‑term morbidity and mortality and measured robustly.

• Function and quality‑of‑life endpoints

• Patient‑reported outcomes (PROs) capturing how patients feel and function (e.g., pain, fatigue, daily activities).

• Performance‑based measures: 6‑minute walk test, functional scales (e.g., NYHA, ECOG), neurocognitive scales.

• For chronic or slowly progressive diseases, these are often co‑primary or key secondary endpoints demonstrating real‑world benefit over time.

2. Surrogate and intermediate endpoints (used cautiously)

Regulators increasingly distinguish between:

• Validated surrogate endpoints

• Defined under U.S. law as markers known to predict clinical benefit, suitable for traditional approval (e.g., LDL‑C reduction for some cardiovascular drugs; blood pressure lowering; certain virologic markers in HIV/hepatitis) [1].

• Listed in FDA’s Surrogate Endpoint Table and updated regularly; continued inclusion is contingent on confirmatory trials sustaining the link to clinical benefit [1].

• “Reasonably likely” surrogate endpoints

• Markers reasonably likely to predict clinical benefit, used for accelerated approval where unmet need is high and waiting for OS or hard outcomes is impractical [1][2].

• In oncology, tumor response rate and PFS have been widely used for accelerated approval, with OS or long‑term outcomes required post‑approval [2][4].

• FDA has explicitly removed surrogates from its list when confirmatory trials failed, reinforcing that surrogates are conditional and context‑specific [1].

• Intermediate/biomarker endpoints in small populations & CGT

• For rare diseases and CGT, FDA’s 2025 draft guidance encourages:

  • Objective, not effort‑dependent measures aligned with disease course in single‑arm and self‑controlled trials (e.g., imaging, lab markers, performance tests) [3].

  • Disease progression modeling with meaningful endpoints correlated to progression (e.g., composite motor scales, organ function markers) [3].

  • Surrogate or intermediate clinical endpoints prior to symptom onset in pre‑symptomatic populations, often supported by digital health technologies (DHTs) to detect subtle functional change [3][4].

3. Endpoints for AI and digital diagnostics

For AI‑driven diagnostic interventions, meaningful endpoints are being reframed around real‑world safety and reliability:

• Misdiagnosis composite as the primary endpoint (rate of missed, delayed, or incorrect diagnoses) rather than just model AUC [5].

• Decision latency (time to actionable decision) and override rates (when clinicians disagree with AI and are ultimately correct) as key secondary endpoints [5].

• Calibration and fairness metrics (e.g., expected calibration error, Brier score, subgroup ΔAUC, ΔFNR) to ensure performance is not only high on average but equitable across demographics [5].

Current trends and evidence

• Regulatory trend toward OS and robust long‑term outcomes:

• Draft guidance on overall survival stresses prespecified OS analyses, event‑driven designs, and full ITT analyses, even when OS is not primary, to capture harm and long‑term benefit [2].

• Emphasis on minimizing missing OS data and appropriate handling of crossover and post‑progression therapies [2].

• Expanding but constrained role of surrogates:

• FDA’s Surrogate Endpoint Table now functions as a curated list of acceptable surrogates, updated every 6 months; endpoints are removed if confirmatory trials fail [1].

• Analyses have shown only about half of accelerated approvals based solely on tumor response ultimately converted to full approval, sharpening scrutiny of such surrogates [4].

• Innovative designs but conventional endpoints:

• Seamless phase II/III and master protocol/platform trials increasingly use short‑term surrogates (e.g., overall response rate) for early decisions, but still rely on OS or PFS as definitive phase III endpoints [4].

• Regulators expect strong control of Type I error, pre‑specified adaptation rules, and simulation‑based operating‑characteristic evaluation in these designs [4].

Counterarguments and nuances

• Surrogates are indispensable in some settings.

• Waiting for OS in indolent cancers, rare diseases, or gene therapies can delay access for years; validated or strongly justified surrogates are often the only feasible path.

• OS is not always practical or uniquely informative.

• In settings with heavy subsequent therapies, extensive crossover, or very long survival, OS becomes confounded or logistically infeasible; PFS, event‑free survival, or composite clinical outcomes may be more interpretable in the short–medium term.

• Patient‑centered endpoints can be noisy and subjective.

• PROs and functional measures vary across cultures and languages and can be influenced by placebo effects; rigorous instrument validation and blinded assessment are necessary.

Practical implications for trial design

• Anchor on a direct clinical endpoint whenever possible, especially for registration:

• OS, event‑free survival, major clinical events, or validated functional outcomes.

• Use surrogates strategically:

• For early decision‑making and accelerated approval, but design credible confirmatory plans linking surrogates to hard outcomes.

• In small/rare and CGT trials:

• Combine self‑controlled or externally controlled designs with objectively measured, disease‑relevant endpoints and progression models that regulators explicitly consider meaningful in context [3][4].

• In AI/diagnostic trials:

• Elevate endpoints such as misdiagnosis rate, clinical decision quality, and fairness across subgroups above pure algorithm metrics [5].